Researchers have developed a personalized AI tutoring system that uses reinforcement learning to adapt the difficulty level of practice problems based on student interactions with a chatbot. The system was tested in a five-month course teaching Python to high school students and found to improve unassisted final exam performance by 6-9 months' worth of schooling, according to mediation analysis. This improvement is attributed to increased engagement, suggesting that the adaptive sequencing algorithm effectively leverages student-chatbot interactions to optimize learning. --- Why it matters: This work matters because it provides large-scale evidence for the effectiveness of using AI in education and demonstrates a novel approach to personalized tutoring that can be applied to various subjects and age groups. The findings have implications for educators and policymakers looking to integrate AI into educational systems. Source: https://arxiv.org/abs/2608.16907